Research On Recommendation System Algorithm Based On Deep Learning

As the number of online movies continues to increase,for movie websites,the high and low scores of movies directly affect the number of views and the choices of users.For users,finding the movie that you really like in a lot of movies is time-consuming and laborious.The recommendation system provides an effective solution to overcome such problems,providing great convenience to websites and users.However,the traditional recommendation algorithm has poor recommendation effect and cold start problem due to data sparseness problem,and does not consider the influence factor of the user's rating on the recommended item when recommending the user.In view of the above problems,this thesis analyzes the recommendation algorithm and its development status,and deeply studies the application of deep learning in the field of film recommendation.By improving the existing algorithms,an interactive hybrid recommendation algorithm is proposed,and the scoring optimization weights are introduced,which effectively alleviates the problem of data sparseness and cold start,and the user's scoring behavior is also optimized.The main research work of this thesis is as follows:(1)Build a scoring optimization algorithm based on collaborative filteringAt present,the recommendation system mainly relies on the similarity calculation to predict the missing values,selects the top N with higher prediction values for recommendation,and does not consider the impact of the user's last movie watching on the scoring behavior.Collaborative filtering scoring optimization algorithm.Based on the collaborative filtering algorithm,the algorithm introduces scoring optimization weights,improves the average score of movies,and reduces the fluctuation of users' ratings on movies.(2)Construct interactive hybrid recommendation algorithm based on selfencoderIn the hybrid recommendation algorithm based on self-encoder,the convergence speed between the deep learning algorithm and the traditional algorithm is inconsistent,which leads to poor recommendation performance.In order to better combine the deep learning with the traditional recommendation algorithm,an interactive mixture based on self-encoder is proposed.Recommended algorithm.The algorithm uses an automatic encoder to perform unsupervised learning on auxiliary information,and performs probability matrix decomposition on the user-scoring matrix.The automatic encoder part and the probability matrix decomposition part interact in the training process,and constrain each other,thereby utilizing the object assistance.Information reconstruction scoring matrix.(3)Proposed the depth coordination bidirectional timing constraint algorithmA deep cooperative bidirectional timing recommendation algorithm is formed by introducing scoring optimization weights in the self-encoder-based interactive hybrid recommendation algorithm.The algorithm consists of a matrix decomposition module,a stack self-encoding module and a score optimization weight module.(4)Test analysisIn this thesis,the test is divided into three parts: a)testing the scoring optimization algorithm based on collaborative filtering,verifying the impact of scoring optimization weight on user scoring behavior;b)mixing recommendation algorithm based on self-encoder and self-encoder based interaction The hybrid recommendation algorithm is tested separately on the Douban movie score dataset and the MovieLens dataset,and the test results are compared and analyzed to verify the impact of the interaction behavior on the recommendation performance.c)The recommended performance of the deep collaborative bidirectional timing constraint algorithm in project recommendation carry out testing.The long-term film recommendation is proposed by the user using the deep cooperative bidirectional timing constraint algorithm.The recommendation results show that the deep cooperative bidirectional timing constraint algorithm effectively alleviates the data sparseness and cold start problem,and also optimizes the user's scoring behavior.